Dafi, Akila Akhtar El (2026) Explainable AI (XAI) untuk Klasifikasi Tingkat Depresi Berbasis EEG pada Terapi Mindfulness Menggunakan Light Gradient Boosting Machine (LightGBM). Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6003251026-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6003251026-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy |
Abstract
Depresi merupakan gangguan kesehatan mental global yang umumnya didiagnosis menggunakan kuesioner subjektif seperti Beck Depression Inventory (BDI). Penelitian ini mengusulkan pendekatan objektif menggunakan sinyal Electroencephalogram (EEG) pada channel Fp1 dan Fp2 untuk klasifikasi tingkat depresi serta mengevaluasi efektivitas Mindfulness-Based Cognitive Therapy (MBCT). Metode klasifikasi yang digunakan adalah Light Gradient Boosting Machine (LightGBM) yang diintegrasikan dengan Explainable AI (XAI) melalui metode Shapley Additive Explanations (SHAP) dan Counterfactual Explanations (CE) untuk interpretabilitas model. Fitur statistik (minimum, maksimum, varians, energi) diekstraksi dari gelombang delta, theta, alpha, dan beta. Hasil penelitian menunjukkan bahwa model LightGBM dengan label hasil Revised Fuzzy C-Means (RFCM) clustering pada skema Stratified K-Fold (K=5) memberikan performa terbaik dengan akurasi 78,5%, spesifisitas 88,5%, dan AUC 82,4%, mengungguli model berbasis label BDI. Analisis SHAP mengungkapkan bahwa fitur varians pada gelombang Beta di area frontal kanan (Fp2) dan kiri (Fp1) merupakan prediktor paling dominan. Analisis CE mengidentifikasi pola pemulihan yang bersifat non-linear. Depresi level 2 ditandai dengan aktivitas otak yang sangat fluktuatif atau tidak stabil, sehingga pemulihan memerlukan proses stabilisasi aktivitas neural. Sebaliknya, depresi level 1 menunjukkan aktivitas otak yang relatif terlalu stabil atau kurang dinamis, sehingga pemulihan menuju kondisi normal memerlukan peningkatan dinamika aktivitas otak agar menjadi lebih responsif dan adaptif. Evaluasi dampak MBCT menunjukkan adanya perbaikan kondisi pada beberapa subjek, namun hasil uji Wilcoxon Signed-Rank Test dan Mann-Whitney U Test menunjukkan bahwa penurunan tingkat depresi pasca-intervensi belum signifikan secara statistik dibandingkan kelompok kontrol, yang mengindikasikan perlunya evaluasi lebih lanjut dengan sampel yang lebih besar.
=====================================================================================================================================
Depression is a global mental health disorder commonly diagnosed using subjective questionnaires such as the Beck Depression Inventory (BDI). This study proposes an objective approach utilizing Electroencephalogram (EEG) signals from Fp1 and Fp2 channels to classify depression levels and evaluate the effectiveness of Mindfulness-Based Cognitive Therapy (MBCT). The classification method employed is the Light Gradient Boosting Machine (LightGBM), integrated with Explainable AI (XAI) through Shapley Additive Explanations (SHAP) and Counterfactual Explanations (CE) to ensure model interpretability. Statistical features (minimum, maximum, variance, energy) were extracted from delta, theta, alpha, and beta waves. The results demonstrated that the LightGBM model using Revised Fuzzy C-Means (RFCM) clustering-derived labels with a Stratified 5-Fold Cross-Validation scheme yielded the best performance, achieving an accuracy of 78.5%, specificity of 88.5%, and an AUC of 82.4%, outperforming models based on BDI labels. SHAP analysis revealed that variance features in Beta waves at the right (Fp2) and left (Fp1) frontal areas were the most dominant predictors. Furthermore, CE analysis reveals a non-linear recovery pattern. Level 2 depression is characterized by highly fluctuating and unstable brain activity, indicating that recovery requires stabilization of neural activity. In contrast, Level 1 depression exhibits relatively overly stable or less dynamic brain activity, suggesting that recovery toward a normal condition requires enhancing neural dynamics to achieve a more responsive and adaptive state. Evaluation of the MBCT impact showed descriptive improvements in 30.4% of subjects. However, results from the Wilcoxon Signed-Rank Test and Mann-Whitney U Test indicated that the post-intervention reduction in depression levels was not yet statistically significant compared to the control group, suggesting the need for further evaluation with a larger sample size.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Depresi, Electronecephalogram, Light Gradient Boosting Machine, Explainable Artificial Intelligence, Mindfullness-Based Cognitive Therapy ============================================================ Depression, Electronecephalogram, Light Gradient Boosting Machine, Explainable Artificial Intelligence, Mindfullness-Based Cognitive Therapy |
| Subjects: | Q Science > Q Science (General) > Q180.55.M38 Mathematical models Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA278 Cluster Analysis. Multivariate analysis. Correspondence analysis (Statistics) |
| Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis |
| Depositing User: | Akila Akhtar El Dafi |
| Date Deposited: | 11 May 2026 01:15 |
| Last Modified: | 11 May 2026 01:15 |
| URI: | http://repository.its.ac.id/id/eprint/133093 |
Actions (login required)
 |
View Item |